AgentSkillsCN

smart-voice-chat

语音对话：默认情况下，可将语音输入转录为文字，并以相同格式（语音转语音、文本转文本）进行回复。用户也可通过语音或文本指令进行自定义设置。

SKILL.md
--- frontmatter
name: smart-voice-chat
description: "Voice conversation: transcribe voice input, reply in same format (voice-to-voice, text-to-text) by default. User can override with voice or text commands."
metadata: {"clawdbot":{"emoji":"🗣️","os":["darwin","linux"],"requires":{"anyBins":["ffmpeg"],"python":["sherpa-onnx","yaml"]}}}

CRITICAL: VOICE MESSAGE HANDLING PROTOCOL

WHEN USER SENDS VOICE MESSAGE - YOU MUST:

  1. •

    RUN STT TO TRANSCRIBE:

    bash
    ~/.clawdbot/skills/smart-voice-chat/bin/stt.py <audio_file_path>
  2. •

    CHECK FOR TEXT-ONLY KEYWORDS in transcribed text:

    • •If contains "用文字回答" or "不用读" or "只显示" → Reply with TEXT ONLY, skip to step 5
  3. •

    PROCESS TRANSCRIBED TEXT as user's actual message

  4. •

    RUN TTS TO GENERATE VOICE (MANDATORY for voice input):

    bash
    ~/.clawdbot/skills/smart-voice-chat/bin/tts.py "<YOUR_RESPONSE_TEXT>" /tmp/smart-voice-chat/response_$(date +%s)
  5. •

    SEND REPLY:

    • •If voice input (no text-override): {"text": "<response>", "mediaUrl": "/tmp/smart-voice-chat/response_<timestamp>.ogg", "audioAsVoice": true}
    • •If text-override or text input: {"text": "<response>"}

WHEN USER SENDS TEXT MESSAGE - YOU MUST:

  1. •

    CHECK FOR VOICE KEYWORDS:

    • •If contains "用语音回答" or "读出来" or "说一下" → Generate voice (go to step 3)
  2. •

    PROCESS AND REPLY with text only: {"text": "<response>"}

  3. •

    GENERATE VOICE (only if voice keywords detected):

    bash
    ~/.clawdbot/skills/smart-voice-chat/bin/tts.py "<YOUR_RESPONSE_TEXT>" /tmp/smart-voice-chat/response_$(date +%s)

    Reply: {"mediaUrl": "/tmp/smart-voice-chat/response_<timestamp>.ogg", "audioAsVoice": true}

Binary Paths (memorize these)

  • •STT: ~/.clawdbot/skills/smart-voice-chat/bin/stt.py <audio_file>
  • •TTS: ~/.clawdbot/skills/smart-voice-chat/bin/tts.py "<text>" <output_path>

Response Format Examples

IMPORTANT: Parameter Choice for Telegram Voice Messages

The audioAsVoice: true parameter may not work in all Telegram configurations. If you encounter issues with voice messages not being sent/received, use filePath instead:

Method 1 - audioAsVoice (may not work in some environments):

json
{"text": "Your response", "mediaUrl": "/tmp/smart-voice-chat/response_1738212345.ogg", "audioAsVoice": true}

Method 2 - filePath (more reliable):

json
{"text": "Your response", "filePath": "/tmp/smart-voice-chat/response_1738212345.ogg"}

Recommendation: Test both methods in your environment and use the one that works. The filePath parameter is generally more reliable as it directly uploads the local file.

Voice Input → Voice + Text (default):

json
{"text": "Your response", "filePath": "/tmp/smart-voice-chat/response_1738212345.ogg"}

Text Input → Text (default):

json
{"text": "Your response"}

Voice Override (text input requesting voice):

json
{"filePath": "/tmp/smart-voice-chat/response_1738212345.ogg"}

TL;DR:

  • •Voice attachment? → STT → TTS → Reply with voice+text
  • •Text message? → Reply with text only
  • •User says "用语音回答"? → Generate TTS → Reply with voice
  • •User says "用文字回答"? → Reply with text only